Accelerating Returns, QES and Guojun Liao's 2026 paper

Guojun Liao has submitted a paper, "Accelerating Returns and the Qualitative Engine for Science," to arXiv on 24 Jun 2026 that offers a mathematical reading of Ray Kurzweil's accelerating-returns thesis and proposes the Qualitative Engine for Science, or QES, as a response to a shortfall in current AI capability. Liao is listed with the Department of Mathematics at The University of Texas at Arlington.

What does Liao say about Kurzweil's accelerating-returns thesis?

Liao gives a simple mathematical interpretation of Kurzweil's central claim that advances across compute, artificial intelligence, brain science, and biotechnology interact so progress becomes self-amplifying and approximately exponential. He argues that acceleration maps most naturally to executional and infrastructural capability, meaning improvements in raw hardware, software pipelines, and scaling behaviors. By contrast, he says, genuine scientific discovery often depends on a different capacity: qualitative reasoning about when a current framework is structurally inadequate and what conceptual move is needed next.

Liao frames this as a distinction between quantitative capability that can accelerate and a qualitative mode of reasoning that does not follow the same dynamics. That distinction drives the paper's central recommendation: accelerating returns alone will not resolve the hard problem of scientific discovery.

How do benchmarks and recent results shape Liao's argument?

Liao cites recent ARC-AGI-3 results to sharpen the distinction he draws: humans solve the benchmark at ceiling, whereas frontier AI systems remain below 1%. He uses that concrete gap to illustrate that, despite strong quantitative progress in some areas, current AI systems still fall far short on tasks where human flexible reasoning performs at ceiling.

The ARC-AGI-3 comparison serves two roles in the paper. First, as empirical evidence that quantitative scaling has not yet reproduced certain human faculties. Second, as a motivation for designing systems aimed at the particular faculty Liao calls qualitative reasoning. The paper links that faculty to the ability to detect structural inadequacy in frameworks and to propose the right conceptual move to progress science.

What is the Qualitative Engine for Science (QES)?

QES is introduced as a response to the missing capacity Liao identifies. He positions QES as an approach focused on organizing and preserving processes of scientific discovery, the kind of human wisdom that includes deciding when a framework is failing and what conceptual change is needed next. Liao emphasizes that QES's value does not depend on when artificial general intelligence arrives; rather, it targets the practices of discovery themselves and aims to make them accessible and organizable.

The paper also cites a cultural and philosophical angle: Demis Hassabis is referenced for emphasizing that humans must retain their sense of meaning and what they choose to focus their lives on. Liao uses that point to underline that the future of AI raises questions about which forms of human understanding are worth preserving and transmitting, and not only how fast compute or models improve.

Why it matters

Liao's distinction reframes a common optimism about exponential progress. If accelerating returns primarily amplify executional and infrastructural capabilities, then scaling alone may produce larger but not necessarily wiser tools. The ARC-AGI-3 data point Liao cites, humans at ceiling versus frontier AI below 1%, gives a concrete measure of that gap. Prioritizing systems or methods that encode qualitative reasoning could shift research attention from pure scale to mechanisms that preserve and extend how humans make conceptual advances.

What to watch

Look for follow-up work that operationalizes QES ideas into testable systems or benchmarks aimed specifically at qualitative reasoning. Also watch whether future ARC-AGI-3 runs or related benchmarks close the gap Liao highlights, or whether new evaluations emerge that measure the kind of framework-detection and conceptual-move abilities QES targets.